Your CLAUDE.md is fighting you.
Let me look at it.
A 90-minute review of your Claude Code setup — file hierarchy, agent rules, session telemetry, and the three habits eating your tokens. You get a written diagnosis, a rewritten root CLAUDE.md, and five templates you can drop into the next project.
Reply within 24h US-Pacific. First three engagements run free as case studies — mention "case study" in the subject if your setup is interestingly broken.
Solo engineer
One-shot, one project, one engineer.
- → 90-min Zoom with screen share, your repo on my screen
- → Written diagnosis delivered within 48h — what's working, what's costing you tokens, what's actively misleading the model
- → One rewritten root
CLAUDE.mdfor the hero project - → Five reusable templates: monorepo, backend service, frontend app, library, scripts
- → One follow-up email round within 7 days
Team of 2–10
Multi-project, multi-engineer, shared memory.
- → Everything in Solo, plus:
- → 2-hour team Zoom — leads + a senior IC each get airtime
- → Audit of how
CLAUDE.mdis shared across repos (monorepo, submodules, copy-paste, or unused) - → Per-project rewrites for up to 3 hero repos
- → Team playbook: who owns the file, when it's reviewed, how new rules graduate from one engineer's session into the shared file
- → Two follow-up email rounds within 14 days
Payments via Stripe invoice after the call. No payment up front, no surprises. Refund anytime within 7 days if the diagnosis doesn't land.
You send me your files
After scheduling, you email me your CLAUDE.md files (root + any nested), a representative session JSONL or two, and a one-paragraph description of what Claude keeps getting wrong.
We meet on Zoom
I share my screen with my notes already pulled up. We walk through the diagnosis live, you ask, I rewrite a section in real time so you see the editorial moves.
You get the deliverables
Within 48 hours: a written diagnosis (PDF or markdown — your pick), the rewritten CLAUDE.md, and the templates as a git-committable folder you can drop into your repos.
I read AI agent memory for a living.
I built AI Memory Reader, the macOS app that browses every memory file ten different AI coding agents leave on disk. To make it work I had to read several gigabytes of real-world CLAUDE.md files, agent rules, and session transcripts — the kind of corpus most engineers never see.
The same patterns show up everywhere. The same five mistakes. The same three habits that quietly burn 30–60% of your token spend with no benefit. I have opinions, and they hold up under load.
I'm a working full-stack engineer — Swift, TypeScript, Go, Python — currently building tools in this space. This isn't a side gig; it's the work.
My repo is private. Will you sign an NDA?
Yes. Mutual NDA before you send anything sensitive — no charge for the NDA round.
I use Codex / Cursor / Aider, not just Claude. Still relevant?
Yes. The same audit covers AGENTS.md, .cursor/rules, and CONVENTIONS.md — the principles are the same and the file syntax is the easy part.
What about a one-off code-review-style audit?
If the deliverable you want is "review my PR" or "review my codebase," that's a different shape of engagement — email me and we'll scope it. The CLAUDE.md audit is intentionally tight.
How fast can we start?
Reply within 24 hours US-Pacific. Most calls happen within a week of the first email.
Are you doing this full-time?
I take 1–2 engagements per week so each one gets real attention. If I'm booked, I'll tell you the next open slot in the first reply.
Can you do this async, without the Zoom?
The Zoom is where the value lands — questions, follow-ups, and live rewrites. If a meeting truly doesn't work, email me anyway and we'll figure it out.
Ready to fix the file that's quietly costing you?
Send a one-paragraph email. I'll reply with availability and the next step.
[email protected]